Abstract: Alpine grassland in the northern Tibet is the largest alpine grassland area of China, The paper aims to accurately obtain the change trend of northern Tibet grassland coverage, using the meteorological data, social statistics data, GIMMS NDVI and MODIS NDVI data as a parameter, to build the BP neural network model, Estimate the trend of annual changes in the grassland in 2010-2014, and use the principal component analysis method to optimize the parameters to improve the model. The result showed that：①The BP neural network model and its improved model for the study area grassland coverage change value and the remote sensing value of correlation coefficient is 0.16, 0.47, It shows that the BP neural network model has good simulation effect after optimizing the parameters through principal component analysis；②The error rate of vegetation index values estimated by two BP neural networks is 2.36% and 2.20%. Both have high simulation accuracy；③From the training steps of neural networks, the training convergence step is 5000 based on the BP neural network model, and the training convergence step length is 454 based on PCA-BP neural network model，It is shown that the latter improves the computational efficiency and shows good convergence，As a result, the annual variation trend fitting degree, vegetation index to estimate the precision value, or from the point of computational efficiency, the improved BP neural network model is more effective to estimate the northern Tibet alpine grassland coverage changes。